Softmax回归的从零实现

#Softmax回归的从零实现
import torch
import torchvision
"""
torchvision是Pytorch中专门用来处理图像的库，这个包中主要包含4大类
torchvision.datasets:是用来进行数据集加载的，帮我们提前处理好了许多图片数据集
torchvision.models:为我们提供了已经训练好的模型，可以加载之后直接使用
torchvision.transforms:该模块提供了一般的图像转换操作类
如:torchvision.transforms.ToTensor()和torchvision.transforms.Normalize()
torchvision.utils:工具类包，负责处理图像
"""
import numpy as np
import sys#该模块提供了Python解释器和一系列运行环境相关的函数
sys.path.append('..')#为了导入上层目录的painting包
import painting as pt

#1.读取和获取数据
batch_size=256
train_iter,test_iter=pt.load_data_fashion_mnist(batch_size)

#2.初始化模型参数
num_inputs=784
num_outputs=10

W=torch.tensor(np.random.normal(0,0.01,(num_inputs,num_outputs)),dtype=torch.float32)
b=torch.zeros(num_outputs,dtype=torch.float32)

#需要对模型参数求梯度
W.requires_grad_(True)
b.requires_grad_(True)

#3.实现softmax运算
def softmax(X):
    X_exp=X.exp()
    partition=X_exp.sum(dim=1,keepdim=True)
    return X_exp/partition #此处应用了广播机制

#4.定义模型
def net(X):
    return softmax(torch.mm(X.view(-1,num_inputs),W)+b)

#5.定义损失函数:交叉熵损失函数
def cross_entropy(y_hat,y):
    return -torch.log(y_hat.gather(1,y.view(-1,1)))

#6.计算分类准确率
def accuracy(y_hat,y):
    return (y_hat.argmax(dim=1)==y).float().mean().item()
#类似地我们可以评价模型net在数据集data_iter上的准确率
def evaluate_accuracy(data_iter,net):
    acc_sum,n=0.0,0
    for X,y in data_iter:
        acc_sum+=(net(X).argmax(dim=1)==y).float().sum().item()
        n+=y.shape[0]
    return acc_sum/n
print(evaluate_accuracy(test_iter,net))

#7.训练模型
num_epochs,lr=5,0.1

#本函数保存在painting中方便以后使用
def train_ch3(net,train_iter,test_iter,loss,num_epochs,batch_size,
              parmas=None,lr=None,optimizer=None):
    for epoch in range(num_epochs):
        train_l_sum,train_acc_sum,n=0.0,0.0,0
        for X,y in train_iter:
            y_hat=net(X)
            l=loss(y_hat,y).sum()

            #梯度清零
            if optimizer is not None:
                optimizer.zero_grad()
            elif parmas is not None and parmas[0].grad is not None:
                for param in parmas:
                    param.grad.data.zero_()
            l.backward()
            if optimizer is None:
                pt.sgd(parmas,lr,batch_size)
            else:
                optimizer.step()#softmax回归的简洁实现中将用到

            train_l_sum+=l.item()
            train_acc_sum+=(y_hat.argmax(dim=1)==y).sum().item()
            n+=y.shape[0]
        test_acc=evaluate_accuracy(test_iter,net)
        print('epoch %d,loss %.4f,train acc %.3f,test acc%.3f'%(epoch+1,train_l_sum/n,train_acc_sum/n,test_acc))

train_ch3(net,train_iter,test_iter,cross_entropy,num_epochs,batch_size,[W,b],lr)